The Bond Market Bought Starlink. The Stock Bought Compute.

SpaceX raised $25 billion in its debut corporate bond sale ^[1], drawing $89 billion in demand, 3.6 times what was on offer, across five tranches maturing as far out as 2056. The sale closed in the same week the company's stock shed $600 billion in market value ^[2]. Two markets looked at the same company and priced different things. The bond market priced Starlink. S&P Global Ratings assigned SpaceX a BBB rating, the lowest investment-grade tier, one notch above speculative, and projected the company will remain cash-flow negative until 2030 ^[2]. That projection means S&P expects SpaceX's compute and AI investments to lose money for at least four more years. The agency endorsed the debt anyway, relying on $18.7 billion in 2025 revenue and more than $100 billion in cash to cover interest payments ^[1][2]. Ark Invest, among the bond buyers, said the existing business segments at their current trajectories are sufficient to justify the investment ^[1]. The proceeds repay a $20 billion bridge loan from the xAI merger and fund general corporate purposes, not compute buildout ^[1]. The equity market priced the compute moonshot. SpaceX's IPO debuted at $135 a share on June 12, a market cap near $2.1 trillion ^[3]. The stock peaked at $225.64 on June 16, pushing the company above $2.99 trillion, larger than Amazon or Microsoft ^[4]. Morningstar cautioned that only the most optimistic scenario, requiring a rapidly reusable Starship and commercially competitive orbital data centers, approaches that price ^[4]. Musk forecast $1 trillion in revenue by 2030 or 2031. Goldman Sachs projected $470 billion. Morgan Stanley, $330 billion ^[5]. Against $18.7 billion in actual 2025 revenue and a $4.9 billion net loss ^[5]. Then the repricing. Between June 18 and 22, SpaceX fell more than 10 percent for three straight sessions, plunging 16.4 percent to $154.60 on June 22 alone ^[2]. The sell-off was not SpaceX-specific. It was part of a broader AI capex repricing that erased $256 billion from Alphabet and $248 billion collectively from Amazon, Meta, and Microsoft, while Accenture fell 18 percent, its worst day ever ^[6]. The FTSE 100 dropped about 100 points overnight on the SpaceX slide, with losses spreading to the same hyperscalers ^[7]. Markets were questioning massive capital expenditures on AI infrastructure ^[6]. SpaceX, priced on exactly those expectations, was caught in the same current. The bond market, priced on what SpaceX already earns, was not. The $25 billion sale closed in that window with $89 billion in orders ^[1]. The divide runs through S&P Global itself. One arm of the company says SpaceX can service its debt. The other says SpaceX is not profitable enough for index inclusion. Neither rests on compute as a revenue engine.

Two S&P divisions, one company

S&P Global Ratings — credit: BBB, lowest investment-grade tier. Projects cash-flow negative until 2030. Endorses $25B in debt on $18.7B revenue and $100B+ cash. ^[2][1]

S&P Dow Jones Indices — index: Blocked fast-track S&P 500 entry. Requires 12-month seasoning and positive GAAP net income. SpaceX's $4.94B loss in 2025 makes it ineligible until at least 2027-2028. Exceptions should not be granted solely based on market capitalization. ^[8]

The compute customers, meanwhile, kept their exits open. Google's deal, $920 million a month from October 2026 through June 2029, roughly $30 billion total, includes a termination clause if SpaceX fails to deliver its full GPU commitment by September 2026 ^[9]. Reflection AI's $6.3 billion lease allows either party to terminate with 90 days' notice after the first three months ^[10]. The revenue the equity market prices as growth is contractually temporary. The debt the bond market holds runs to 2056. Even the timing of skepticism cut against the compute vision. Masayoshi Son, whose SoftBank has committed $65 billion to OpenAI, publicly rejected orbital data centers on June 23, the day SpaceX's sell-off deepened ^[11]. Son argued that electricity savings from space-based solar are offset by hardware, transportation, maintenance, and communication latency, and that the next few years matter more than what might happen a decade out ^[11]. None of this means the bond market is right and the equity market wrong. The equity market's post-plunge floor, a market cap still near $2 trillion ^[12], may still embed compute expectations that Morningstar considers unlikely ^[4]. And the bond market's appetite for AI-infrastructure debt is broad, not SpaceX-specific. Nvidia raised $25 billion in bonds one week earlier under comparable terms, also upsized from $20 billion, also investment-grade, also maturities to 2056, but with $50 billion in cash and proven AI cash flow ^[13]. The IMF and Bank of England have warned that debt-funded AI infrastructure could create systemic risks similar to the dot-com bubble ^[13]. The bond market may be underpricing risk, not endorsing permanence. But the divergence is the point. In the same week, two markets looked at one company. The equity market repriced the compute moonshot, a $600 billion swing driven by AI capex anxiety that hit every hyperscaler. The bond market's assessment held firm: $89 billion in demand for debt backed by Starlink's revenue and a cash pile big enough to absorb years of compute losses. The stock market bought what SpaceX might become. The bond market bought what SpaceX already is.